UCSD Hao AI Lab 将 GPT-6 Astra 与 Codex 机器人智能体提速约 3 倍
Why don't benchmarks show this? In RoboDojo, the simulator waits for the model before it steps, but a real robot doesn't wait.
AI 导读
UCSD Hao AI Lab 通过合并工具调用往返、精简上下文以及让模型推理与机器人运动并行,将受测机器人智能体提速约 3 倍。杯子任务从 11.1 min 缩短至 3.4 min,毛巾折叠从约 18 min 缩短至约 6 min,机器人运动时间占比由 24% 提升至 84%。不过每个动作仍基于约 5 s 前的观测,环境变化后的反应延迟约为 5 s。
正文
Why don't benchmarks show this? In RoboDojo, the simulator waits for the model before it steps, but a real robot doesn't wait.
Our early towel-folding runs took 13 to 37 minutes. In one of the longest, the model took 78% of the wall time, and the robot moved for only a quarter of the run.
AI 热点